[CUDA][HIP] Fix hostness check with -fopenmp

CUDA/HIP determines whether a function can be called based on
the device/host attributes of callee and caller. Clang assumes the
caller is CurContext. This is correct in most cases, however, it is
not correct in OpenMP parallel region when CUDA/HIP program
is compiled with -fopenmp. This causes incorrect overloading
resolution and missed diagnostics.

To get the correct caller, clang needs to chase the parent chain
of DeclContext starting from CurContext until a function decl
or a lambda decl is reached. Sema API is adapted to achieve that
and used to determine the caller in hostness check.

Reviewed by: Artem Belevich, Richard Smith

Differential Revision: https://reviews.llvm.org/D121765

1 files changed, 10 insertions, 10 deletions

diff --git a/clang/lib/Sema/SemaCUDA.cpp b/clang/lib/Sema/SemaCUDA.cpp
index 9278551..b0af130 100644
--- a/clang/lib/Sema/SemaCUDA.cpp
+++ b/clang/lib/Sema/SemaCUDA.cpp
@@ -728,8 +728,9 @@ void Sema::MaybeAddCUDAConstantAttr(VarDecl *VD) {
 Sema::SemaDiagnosticBuilder Sema::CUDADiagIfDeviceCode(SourceLocation Loc,
                                                        unsigned DiagID) {
   assert(getLangOpts().CUDA && "Should only be called during CUDA compilation");
+  FunctionDecl *CurFunContext = getCurFunctionDecl(/*AllowLambda=*/true);
   SemaDiagnosticBuilder::Kind DiagKind = [&] {
-    if (!isa<FunctionDecl>(CurContext))
+    if (!CurFunContext)
       return SemaDiagnosticBuilder::K_Nop;
     switch (CurrentCUDATarget()) {
     case CFT_Global:
@@ -743,7 +744,7 @@ Sema::SemaDiagnosticBuilder Sema::CUDADiagIfDeviceCode(SourceLocation Loc,
         return SemaDiagnosticBuilder::K_Nop;
       if (IsLastErrorImmediate && Diags.getDiagnosticIDs()->isBuiltinNote(DiagID))
         return SemaDiagnosticBuilder::K_Immediate;
-      return (getEmissionStatus(cast<FunctionDecl>(CurContext)) ==
+      return (getEmissionStatus(CurFunContext) ==
               FunctionEmissionStatus::Emitted)
                  ? SemaDiagnosticBuilder::K_ImmediateWithCallStack
                  : SemaDiagnosticBuilder::K_Deferred;
@@ -751,15 +752,15 @@ Sema::SemaDiagnosticBuilder Sema::CUDADiagIfDeviceCode(SourceLocation Loc,
       return SemaDiagnosticBuilder::K_Nop;
     }
   }();
-  return SemaDiagnosticBuilder(DiagKind, Loc, DiagID,
-                               dyn_cast<FunctionDecl>(CurContext), *this);
+  return SemaDiagnosticBuilder(DiagKind, Loc, DiagID, CurFunContext, *this);
 }
 
 Sema::SemaDiagnosticBuilder Sema::CUDADiagIfHostCode(SourceLocation Loc,
                                                      unsigned DiagID) {
   assert(getLangOpts().CUDA && "Should only be called during CUDA compilation");
+  FunctionDecl *CurFunContext = getCurFunctionDecl(/*AllowLambda=*/true);
   SemaDiagnosticBuilder::Kind DiagKind = [&] {
-    if (!isa<FunctionDecl>(CurContext))
+    if (!CurFunContext)
       return SemaDiagnosticBuilder::K_Nop;
     switch (CurrentCUDATarget()) {
     case CFT_Host:
@@ -772,7 +773,7 @@ Sema::SemaDiagnosticBuilder Sema::CUDADiagIfHostCode(SourceLocation Loc,
         return SemaDiagnosticBuilder::K_Nop;
       if (IsLastErrorImmediate && Diags.getDiagnosticIDs()->isBuiltinNote(DiagID))
         return SemaDiagnosticBuilder::K_Immediate;
-      return (getEmissionStatus(cast<FunctionDecl>(CurContext)) ==
+      return (getEmissionStatus(CurFunContext) ==
               FunctionEmissionStatus::Emitted)
                  ? SemaDiagnosticBuilder::K_ImmediateWithCallStack
                  : SemaDiagnosticBuilder::K_Deferred;
@@ -780,8 +781,7 @@ Sema::SemaDiagnosticBuilder Sema::CUDADiagIfHostCode(SourceLocation Loc,
       return SemaDiagnosticBuilder::K_Nop;
     }
   }();
-  return SemaDiagnosticBuilder(DiagKind, Loc, DiagID,
-                               dyn_cast<FunctionDecl>(CurContext), *this);
+  return SemaDiagnosticBuilder(DiagKind, Loc, DiagID, CurFunContext, *this);
 }
 
 bool Sema::CheckCUDACall(SourceLocation Loc, FunctionDecl *Callee) {
@@ -794,7 +794,7 @@ bool Sema::CheckCUDACall(SourceLocation Loc, FunctionDecl *Callee) {
 
   // FIXME: Is bailing out early correct here?  Should we instead assume that
   // the caller is a global initializer?
-  FunctionDecl *Caller = dyn_cast<FunctionDecl>(CurContext);
+  FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true);
   if (!Caller)
     return true;
 
@@ -860,7 +860,7 @@ void Sema::CUDACheckLambdaCapture(CXXMethodDecl *Callee,
 
   // File-scope lambda can only do init captures for global variables, which
   // results in passing by value for these global variables.
-  FunctionDecl *Caller = dyn_cast<FunctionDecl>(CurContext);
+  FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true);
   if (!Caller)
     return;